Make the amount of IO parallelism configurable (i.e. number of separate
index sections to be processed in parallel).

Defining PARALLEL_IO = 20 restores approximately the IO pattern of the
original preload_index() implementation. Best performance is achieved with
PARALLEL_IO = 1 (i.e. sequential IO).

Signed-off-by: Karsten Blees <bl...@dcon.de>
---

Applies on top of "preload-index: optimize for sequential IO".

 preload-index.c | 22 +++++++++++++++++-----
 1 file changed, 17 insertions(+), 5 deletions(-)

diff --git a/preload-index.c b/preload-index.c
index 6ac368d..5fe5521 100644
--- a/preload-index.c
+++ b/preload-index.c
@@ -23,6 +23,7 @@ static void preload_index(struct index_state *index,
  */
 #define MAX_PARALLEL (20)
 #define THREAD_COST (500)
+#define PARALLEL_IO (1)
 
 struct thread_data {
        pthread_t pthread;
@@ -38,23 +39,34 @@ static void *preload_thread(void *_data)
        struct thread_data *p = _data;
        struct index_state *index = p->index;
        struct cache_def cache;
+       int blocks = DIV_ROUND_UP(index->cache_nr, THREAD_COST);
+       int step = DIV_ROUND_UP(blocks, PARALLEL_IO);
 
        memset(&cache, 0, sizeof(cache));
        for (;;) {
                /* get next batch of entries to check */
                pthread_mutex_lock(p->pmutex);
                nr = *p->pnr;
-               *p->pnr += THREAD_COST;
+               (*p->pnr)++;
                pthread_mutex_unlock(p->pmutex);
 
+               /* break loop if no more work to do */
+               if (nr >= blocks + PARALLEL_IO - 1)
+                       break;
+
+               /*
+                * rearrange iteration order, i.e. with PARALLEL_IO = 2,
+                * [0, 1, 2, 3, 4, 5, 6, 7] becomes [0, 4, 1, 5, 2, 6, 3, 7]
+                */
+               nr = (nr / PARALLEL_IO) + (nr % PARALLEL_IO) * step;
+               if (nr >= blocks)
+                       continue;
+               nr *= THREAD_COST;
+
                max_nr = nr + THREAD_COST;
                if (max_nr > index->cache_nr)
                        max_nr = index->cache_nr;
 
-               /* break loop if no more work to do */
-               if (nr >= max_nr)
-                       break;
-
                for (; nr < max_nr; nr++) {
                        struct cache_entry *ce = index->cache[nr];
                        struct stat st;
-- 
1.9.4.msysgit.0.1.gc8a51b4

--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to